Skip to content
Articles
ArticlesAI translation· 10 November 2025

Silicon Valley's AI moat has a leak, and it's called open source

by 4 min read

The myth of the impregnable fortress

In Silicon Valley's strategy suites, people love telling the story of the impregnable moats. The AI race, so the legend goes, is a game for giants with budgets the size of small nations. Only a handful of US tech companies can play; the rest of the world watches in awe. A lovely, reassuring story. The only problem: while the giants gaze smugly at their walls, a little boat from China is crossing the supposedly impassable moat as if it were a creek. That should make for a few restless nights in the C-suite.


When open source changes the rules of the game

The old rule of thumb is dead: proprietary models define the cutting edge, while the open-source community respectfully tries to close the gap. That dogma no longer holds, and the challenge is coming squarely from China. After Deepseek got the ball rolling, Moonshot AI is now escalating things with Kimi K2 Thinking, an open-weights model that takes on the leading proprietary systems head-on. Open source, by now coming mostly out of Chinese labs, is no longer just the chaser. It is now an equal competitor at the very front.

The raw performance numbers speak plainly and should give everyone pause:

  • Artificial Analysis Intelligence Index: 67 (the highest score ever recorded for an open-weights model)
  • HLE benchmark (no tools): 22.3% (new record for core knowledge)
  • HLE benchmark (as an AI agent): 44.9% (new record)
  • BrowseComp (as an AI agent): 60.2% (new record)

This model wasn't simply stuffed with knowledge. It was deliberately designed as a “thinking agent.” It can coherently execute up to 200–300 sequential tool calls without a human stepping in. That's a big leap for a freely available model.

But the real strategic bombshell lies less in the performance itself than in how it's being distributed. Kimi K2 is released under a modified MIT license. Translation: practically unrestricted commercial use for anyone. The only condition, for services with more than 100 million monthly users, is to display the name “Kimi K2” prominently. In other words: if you build a compact car, you can just use it; if you build a Formula 1 car, you have to put the “sponsor” sticker on it.

And that's exactly where the moat is leaking. The point isn't merely that a powerful model exists; it's that Moonshot is more or less giving it away to the world. This permissive license is the delivery mechanism that turns raw AI intelligence into a freely available commodity and forces the competition to differentiate on entirely new battlefields.


Intelligence (?) becomes a commodity

The release of Kimi K2 is yet another strategic wake-up call. It proves that raw AI is increasingly becoming an interchangeable commodity.

The $5 million wake-up call

Kimi K2 was reportedly developed on a training budget of under 5 million dollars. To put that in perspective: in Silicon Valley, that kind of money might pass as the “catering budget” for a bigger project. This fact pulverizes the narrative that top-tier performance is reserved exclusively for those juggling billions.

AI's new battlefields

For strategic decision-makers, three unmistakable takeaways follow:

The performance gap between open-source and proprietary AI is closing. It was never that big to begin with.
Episode 10.9 - AI is not a product, it's a featureDocIsInDaHouse
Benchmark wins are irrelevant if operational efficiency doesn't add up. Coz' it's the process, stupid!
Episode 10.6 - Generate or delegate?DocIsInDaHouse
Chinese AI labs have become central players at the very front of AI development.

For the established US providers, this means their competitive edge is eroding fast, as expected. Future success will depend less on raw model performance. The new decisive differentiators are reliability, superior user experience (UX) and easy product integrations.


Europe's opportunity: a blueprint for digital sovereignty?

While the US should be pondering its shrinking advantages, Europe should be paying close attention. The success of Chinese open-source AI is a threat to US dominance, yes, and at the same time a strategic blueprint for Europe's own ambitions for digital sovereignty. For European strategists, these are the decisive levers to pull now. China's open-source blueprint hands Europe a ready-made playbook with five strategic advantages:

  • Transparency and auditability: Open source code enables independent review and ensures compliance with European ethical standards such as the EU AI Act.
  • Independence and control: Open source prevents vendor lock-in by US corporations and allows adaptation to local needs without crushing licensing models.
  • Fostering innovation and participation: The barriers for research, SMEs and educational institutions come down, which strengthens the entire ecosystem.
  • Strengthening European infrastructure: Open-source models are the foundation for initiatives such as GAIA-X, EURO-LLM and OpenGPT-X, which put European standards at the center.
  • Regulatory fit: The transparency and control required by the EU AI Act, especially for high-risk systems, are far easier to deliver with open-source approaches.

The facts are on the table. A small Chinese AI lab, with a fraction of the budget, has delivered a model that challenges the expensive, closed systems of the US giants, and then hands it to the world for free on top of that. For Europe, this is an open invitation to use the rules of the game to its own advantage. So the decisive question is no longer whether, but how fast European players pick up on this inspiration.

Or do we just want to keep watching while others shape our future?

Live long and prosper 😉✋

Read on
Topic
AI Technology
16 PostsView all →
▶Listen